Clean up local launch wording

Author: hasandemirkiran

README.md

@@ -36,7 +36,7 @@ kontext start
 
 This starts the currently supported adapter, Claude Code, with a local Kontext runtime. No hosted login is required.
 
-By default, Kontext runs in observe mode: the agent keeps running, while Kontext records `would allow`, `would ask`, and `would deny` decisions in the local dashboard. The dashboard is served on loopback, with the URL printed at startup.
+By default, Kontext runs in observe mode: the agent keeps running, while Kontext records `would allow` and `would deny` decisions in the local dashboard. The dashboard is served on loopback, with the URL printed at startup.
 
 To block supported risky pre-tool actions, start in enforce mode:
 
@@ -46,11 +46,11 @@ KONTEXT_MODE=enforce kontext start
 
 ## Core features
 
-Kontext balances security and utility for AI agents: low-risk actions keep moving, sensitive actions require approval, and unsafe actions can be blocked before they execute.
+Kontext balances security and utility for AI agents: low-risk actions keep moving, and unsafe actions can be blocked before they execute.
 
 - **Audit trails:** Record who instructed which agent to do what, what the agent accessed, which tools it called, what policy decisions were made, and what happened next. Build a chain of custody for security review, incident investigation, and compliance evidence.
-- **Deterministic policy:** Apply `allow`, `ask`, and `deny` rules to agent actions at runtime, before they execute. Use hard policies for known boundaries such as destructive commands, production resources, sensitive files, data exports, and credential access.
-- **Probabilistic risk detection:** Route actions that deterministic policy allows through a local judge for an additional allow/deny decision without sending tool context to hosted services.
+- **Deterministic policy:** Apply `allow` and `deny` rules to agent actions at runtime, before they execute. Use hard policies for known boundaries such as destructive commands, production resources, sensitive files, data exports, and credential access.
+- **Local LLM judge:** Route actions that deterministic policy allows through a local judge for an additional allow/deny decision without sending tool context to hosted services.
 - **Credential injection:** Inject scoped OAuth credentials at runtime using RFC 8693-compliant OAuth 2.0 Token Exchange, so agents can access approved tools without users pasting secrets into chat, config files, or project environments. Credentials can be short-lived, least-privilege, and bound to the current user, session, or workflow.
 
 The local decision path is:
@@ -61,8 +61,8 @@ Agent tool call
   -> local runtime socket
   -> action classification
   -> deterministic policy
-  -> probabilistic risk score
-  -> allow / ask / deny
+  -> local LLM judge
+  -> allow / deny
   -> local dashboard
 ```
 
@@ -90,7 +90,7 @@ For enterprise identity, audit retention, organization controls, deployment plan
 | Default | Behavior |
 | --- | --- |
 | Local evaluation | Default `kontext start` does not require hosted login or trace upload. |
-| Observe mode | Decisions are recorded as `would allow`, `would ask`, or `would deny` without blocking the agent. |
+| Observe mode | Decisions are recorded as `would allow` or `would deny` without blocking the agent. |
 | Loopback dashboard | The local dashboard binds to loopback by default. |
 | Redacted storage | Tool events and decisions are stored locally with redaction. |
 | No reasoning capture | Kontext captures tool events and outcomes, not LLM reasoning, token usage, or full conversation history. |
@@ -118,7 +118,7 @@ kontext start
   |-- Local runtime: Unix socket service + RuntimeCore
   |-- Local dashboard: 127.0.0.1:4765
   |-- Deterministic policy: curated rule categories + active profile
-  |-- Probabilistic risk: score + ask escalation for ambiguous actions
+  |-- Local LLM judge: localhost allow/deny decision after deterministic allow
   |-- Store: local SQLite with redacted events and decision metadata
 ```
 

docs/guard.md

@@ -2,21 +2,19 @@
 
 Guard is the local safety mode inside `kontext`.
 
-It lets a developer run Claude Code normally while Kontext watches tool calls locally, redacts captured data, stores events in local SQLite, and shows a local dashboard with `would allow`, `would ask`, and `would deny` decisions.
+It lets a developer run Claude Code normally while Kontext watches tool calls locally, redacts captured data, stores events in local SQLite, and shows a local dashboard with `would allow` and `would deny` decisions.
 
 ## User path
 
 ```bash
 brew install kontext-security/tap/kontext
-kontext guard start
-claude
+kontext start
 ```
 
 Until the Guard PR is merged and released, test from source:
 
 ```bash
-go run ./cmd/kontext guard start
-claude
+go run ./cmd/kontext start
 ```
 
 ## Runtime boundary
@@ -31,10 +29,10 @@ Guard mode is local-first by default:
 - embedded local dashboard
 - observe mode by default
 
-Hosted mode remains separate:
+Hosted managed mode remains separate:
 
 ```bash
-kontext start --agent claude
+kontext start --managed --agent claude
 ```
 
 Hosted mode owns login, provider connection, short-lived scoped credentials, hosted traces, and team governance.
@@ -61,7 +59,7 @@ Guard uses two layers:
 
 ## Local judge
 
-Guard can optionally call a localhost OpenAI-compatible judge, such as `llama-server`, after deterministic rules allow a blocking tool call:
+The user-facing `kontext start` path manages a local judge by default. For daemon-only diagnostics, Guard can call a localhost OpenAI-compatible judge, such as `llama-server`, after deterministic rules allow a blocking tool call:
 
 ```bash
 kontext guard start \

internal/guard/cli/cli.go

@@ -172,7 +172,7 @@ func runDaemon(ctx context.Context, args []string, out io.Writer) error {
 	defer runtimeService.Stop()
 	fmt.Fprintf(out, "Kontext Guard local daemon listening on http://%s\n", *addr)
 	fmt.Fprintf(out, "Hook runtime: unix://%s\n", *socketPath)
-	fmt.Fprintln(out, "Mode: observe (Claude Code runs normally; decisions are recorded as would allow / would ask / would deny).")
+	fmt.Fprintln(out, "Mode: observe (Claude Code runs normally; decisions are recorded as would allow / would deny).")
 	fmt.Fprintf(out, "Dashboard: http://%s\n", *addr)
 	if localJudge != nil {
 		fmt.Fprintf(out, "Local judge: %s\n", judgeStatus)
@@ -540,7 +540,7 @@ func installClaudeHooks(out io.Writer, socketPath string) error {
 	}
 	fmt.Fprintf(out, "Installed Kontext Guard Claude Code hooks into %s\n", settingsPath)
 	fmt.Fprintf(out, "Hook command: %s\n", hookCommand)
-	fmt.Fprintln(out, "Default mode is observe. Set KONTEXT_MODE=enforce later to block ask/deny decisions.")
+	fmt.Fprintln(out, "Default mode is observe. Set KONTEXT_MODE=enforce later to block deny decisions.")
 	return nil
 }
 

internal/run/local.go

@@ -75,8 +75,8 @@ func StartLocal(ctx context.Context, opts Options) error {
 
 func printLocalMode(out *os.File, mode string) {
 	if mode == "enforce" {
-		fmt.Fprintln(out, "Mode: enforce (ask and deny decisions block supported pre-tool actions).")
+		fmt.Fprintln(out, "Mode: enforce (deny decisions block supported pre-tool actions).")
 		return
 	}
-	fmt.Fprintln(out, "Mode: observe (agent actions run normally; decisions are recorded as would allow / would ask / would deny).")
+	fmt.Fprintln(out, "Mode: observe (agent actions run normally; decisions are recorded as would allow / would deny).")
 }